|
Abstract:
|
Observational studies of social phenomena often use non-randomized samples, so it is unclear whether their results can be generalized to a larger population. We build upon Lorenz et al. (2006) which found that divorced women in the Iowa Midlife Transitions (MT) Project reported more illness 10 years after divorce than their married counterparts. We construct a basic model to compare three potential mediating factors, including poorer work conditions, increased social isolation, and poorer health care, and evaluate the sensitivity of results to several hypothetical sampling scenarios. Under these scenarios, we evaluate alternative weighting and estimation schemes for the data, and assess the robustness of the findings. We argue that results which hold across multiple weighting schemes are likely to be valid more generally, despite the initial non-random selection of study respondents.
|
